Financial Liberalization and Consumption Volatility in Developing Countries Author(s): Andrei A. Levchenko Source: IMF Staff Papers, Vol. 52, No. 2 (2005), pp. 237-259 Published by: Palgrave Macmillan Journals on behalf of the International Monetary Fund Stable URL: http://www.jstor.org/stable/30035897 . Accessed: 25/06/2014 05:06 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp

. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

.

Palgrave Macmillan Journals and International Monetary Fund are collaborating with JSTOR to digitize, preserve and extend access to IMF Staff Papers.

http://www.jstor.org

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

IMFStaffPapers Vol. 52, Number2 C 2005 International MonetaryFund

FinancialLiberalization and ConsumptionVolatility in Developing Countries ANDREIA. LEVCHENKO* One of the chief benefitsoffinancial liberalizationproposedby theoreticalliterature is that it shouldallow countriesto bettersmoothconsumptionthroughinternational risk sharing. Recent empirical evidence does not support this prediction. In developing countries, financial liberalization seems to be associated with an increase in consumption volatility. This paper seeks to rationalize the evidence by linking it to two important features of developing countries. First, domestic financial markets are underdeveloped. We model this by adopting the Kocherlakota (1996) framework of risk sharing subject to limited commitment. Second, access to international markets is not available to all members of society. We show that when risks are idiosyncratic, that is, insurable within the domestic economy, opening up to international markets reduces the amount of risk sharing attained at home and raises the volatility of consumption. When risk is aggregate to the economy, the underdevelopedfinancial system prevents the pooling of aggregate risk across agents for the purposes of insurance in the international markets. Thus, while the volatility of consumption coming from aggregate risk decreases with financial liberalization, it does so by much less than would be predicted by a representative agent model. [JEL F02, F36]

ow

does internationalfinancial integrationhelp developing countries?Two main potentialbenefits are (i) the more efficient allocation of capital across

ResearchDepartment, International Fund.Theauthoris gratefulto Daron *Economist, Monetary Acemoglu,MichaelAlexeev, FernandoAlvarez,MariosAngeletos,Olivier Blanchard,Bjoern JaumeVentura, ShawnCole,SimonJohnson, AyhanKose,PaoloMauro,PetiaTopalova, Bruegemann, atMIT,andparticipants of theIMFAnnualResearch Wei,IvanWerning, Shang-Jin workshop participants forhelpfulcomments. Financial fromtheDavid Conference andtheLACEA AnnualConference support FinchFoundationis gratefullyacknowledged.

237

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA. Levchenko countryborders,and (ii) improved risk-sharingopportunities.For risk sharingin particular,internationalfinancial integration should lead to a decrease in consumption volatility relative to output volatility. Indeed, the last few decades saw ever-increasingcapitalflows across nationalborders.Whatdid we learn aboutthe effects of internationalfinancial integrationon the volatility of consumption? The latest empirical evidence suggests that the outcomes are quite different from those predictedby the conventional risk sharingmodels. Kose, Prasad,and Terrones (2003) examine the volatility of consumption relative to income for a broadsample of developingcountries.The resultsare quite puzzling. Consumption volatility relative to income volatility increasedbetween the 1980s and the 1990s for the more financially integrateddeveloping countries.The period in questionis precisely the time of increasedcross-bordercapitalflows thatshouldhave afforded those countriesan opportunityto smoothconsumptionin the face of income shocks. The authorspoint out that these results cannot be explained away by some countries experiencing crises, because they look at consumptionvolatility relative to that of income. The regressionanalysis corroboratesthese results.Financialopenness, measuredby the gross capital flows relative to GDP, is associated with an increase in the ratio of consumption volatility to income volatility, up to a certain level of financialopenness. Beyond that level, financialintegrationseems to lower consumptionvolatility. The main purposeof this paperis to explore an explanationfor the perplexing empiricalevidence. We study the effects of financialliberalizationon developing countriesin light of two importantfeaturesof these countries.First,domestic institutionsandfinancialmarketsareunderdeveloped.Second,not all agentshave access to the internationalfinancial markets.In this framework,we reach three main results. When risks are purely idiosyncratic,that is, perfectly insurablewithin the domestic economy, opening up to internationalmarketsreducesthe amountof risk sharing attained at home and raises the volatility of consumption. When risk is purelyaggregateto the economy,the underdevelopedfinancialsystem preventsthe pooling of aggregaterisk across agents for the purposesof insurancein the internationalmarkets.Thus, while the volatility of consumptiondecreases with opening in this case, it does so by much less than in a frictionless model. Finally, the gains from financial integrationare unevenly distributed.Agents who have direct access to internationalmarketsbenefit disproportionately,while those who do not may actually experience an increase in their consumptionvolatility and a decline in welfare. This paper representsa very differenttreatmentof the relationshipbetween financialintegrationand consumptionvolatility.In thinkingaboutthis relationship, our intuitionsare typicallyshapedby representativeagent models such as Obstfeld (1994) or ObstfeldandRogoff (2000). By construction,these models can tell us only aboutthe role of financialintegrationin sharingaggregatecountryrisk. The representativeagent models make two implicit assumptions.First,to the extent there is idiosyncraticrisk among agents within a country,these agents reach the first best level of risk sharing,and only aggregaterisk remainsto be insuredabroad.Second, the aggregatecountry risk is perfectly pooled across agents, or, alternatively,all agents have equal access to the internationalmarkets. 238

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIALLIBERALIZATION AND CONSUMPTIONVOLATILITY

To help rationalizethe disconcertingempiricalevidence, this paperfocuses on precisely the aspectsmissing from the traditionalanalysis.We move away from the representativeagent framework.In our model, agents are heterogeneousboth in their income process and in whetherthey have access to the internationalfinancial markets.Of course,this approachis fruitfulonly when the within-countryrisk-sharing arrangementis subjectto frictions,but we believe thatthese frictionsare important,especially in the less developed countries.Indeed,the hypothesisof complete consumptionrisk sharingis typicallyrejectedeven in economies with highly sophisticatedfinancialmarketssuch as the United States (Attanasioand Davis, 1996; and Hayashi,Antonji,and Kotlikoff,1996). In developingcountrieswith poor qualityof contractinginstitutions,obstacles to sharing idiosyncraticconsumptionrisks are boundto be even more severe. The basic model is a version of the Kocherlakota(1996) or Kehoe and Levine (2001) frameworkof risk sharingsubjectto limitedcommitment.In the model there aretwo groupsof people whose income processesmay differ.They enterinto a risksharingrelationship,subjectto the constraintthat participationby each agent must be voluntaryin all datesand states.We view this constraintas a consequenceof poor contractenforcementand an underdevelopedfinancial system. Agents cannot successfully committheirfutureincome flows to the risk-sharingrelationship. The voluntaryparticipationconstraintmeansthatthe firstbest level of risk sharing is not necessarilyachieved.Agents with high currentincome realizationswill be temptedto walk away from the risk-sharingarrangementandenjoy the high current consumption.If the agentdoes walk away,however,the risk-sharingarrangementis severedforever.Naturally,each agent'soutsideoptionwill be key in determiningthe extent to which voluntaryparticipationis sustainable.If the outside option is very good, the risk-sharingrelationshipmay not be viable, becausethe agentwill choose to walk away the firsttime her currentincome shock is high. Since financialopening will affect some agents'outsideoptions,it will have an importanteffect on the state of domesticrisk sharing. We model financial opening as allowing only one group of agents access to internationalmarkets.We call these agentsthe upper,or middle,class. The assumption thatonly some groupswill have access to foreign marketsseems plausiblefor developing countries.For simplicity,we will think of the foreign marketsas providing an exogenous amountof insurance,and we do not model them explicitly. Whenthe upperclass gains access to the internationalmarkets,it chooses the amount of its participationin the domestic and foreign marketsoptimally.Thus, we extend the basic Kocherlakotaframeworkto endogenize the extent of participationin the domestic risk-sharingarrangementby one of the groups. What effects will financial liberalizationhave in this economy? We consider two polar cases. First, suppose that the groups face purely idiosyncraticrisks and thataggregatecountryrisk is absent.In the frictionlessbenchmark,thereis no insurance role for the internationalmarkets.Whendomesticrisk sharingis subjectto frictions, however,access to internationalmarketshas importantconsequencesthrough its effect on agents' outside options. As the upper class experiences a dramatic increase in their outside option, the extent of risk sharingattainablein the domestic relationshipis reduced.Furthermore,the less attractivethe domesticrisk-sharing 239

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA. Levchenko relationshipbecomes, the more likely it is that the upperclass will reduce its participation in it and insure abroadinstead.When they do so, the agents left behind in the domestic risk-sharingrelationshipexperience an increase in consumption volatility,because the income from the upperclass is no longer availableto insure them. Thus, when access to internationalfinancial marketsis quite uneven in the economy, some groups' participationin these marketsactually lowers the extent of risk sharing available at home. As a result, the members of society unable to take advantageof internationalfinancialintegrationwill be adverselyaffected,and their consumptionvolatilitywill increase.It is importantto note that in this model, agents use internationalmarketsto insure idiosyncraticrisks, that is, risks that are in principleinsurablewithin the domestic economy. This possibilityhas not, to our knowledge, been consideredin the literatureso far. The second polar case we consider is that of only aggregateuncertainty.All agents in the economy face the same income process, but they are nonethelessheterogeneousin whetherthey have access to internationalmarkets.In the frictionless benchmark,it is not importantwhich of the agentshave access to internationalmarkets, because the aggregateincome risk would be pooled and insuredabroadoptimally by the agents able to do so. In our framework,the voluntary-participation constraintin the domesticmarketspreventsthis fromoccurring.The upperclass will certainlyinsure its income shocks abroad.However, there are limits to how much of the lower class's income it can insure. We show that in this framework,even aggregatecountryrisk may not be fully eliminated. It is also clear that the benefits from financial integration are unevenly distributedbetween the groups, with the upper class enjoying them fully, while the lower class benefits less. Furthermore,when idiosyncratic risks predominate in the economy, agents who do not have access to internationalmarkets may actually lose from financial liberalization, as their opportunities for insuring income risk decrease. A largeliteraturestudiesthe relationshipbetweenfinancialand tradeintegration and output volatility. Since integrationaffects agents' investment and asset allocationdecisions, it naturallychanges the volatility of output.A numberof very differentmodels, such as BaxterandCrucini(1995) andAcemogluandZilibotti(1997) show that financial integrationmay increaseoutputvolatility.An importantstrand of the literatureanalyzes the role of speculative capital flows in precipitating financial crises in emerging markets,with importantconsequences for real output volatility. Kaminsky and Reinhart(1999) argue that emerging marketcrises are frequentlytwin crises, in which a balance of payments crisis is combined with a banking crisis. In this framework,foreign capital inflows prone to sudden stops i la Calvo (1998) exacerbatedistortionsin the domesticbankingsystem andincrease the likelihood of crises. This paperaddressesa differentquestion. Suppose that internationalfinancial integrationdoes increase outputvolatility in emerging markets,although,perhaps surprisingly,this does not appearto be the case (see Kose, Prasad,and Terrones, 2003). A representativeagent model in which agents can use internationalmarkets to insureagainstoutputrisk would still imply thatconsumptionvolatility,and certainly the ratioof consumptionvolatility to income volatility,should decline under 240

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AND CONSUMPTION FINANCIAL LIBERALIZATION VOLATILITY quite general conditions.' Here, we provide a frameworkthat shows that domestic frictionsand uneven access to internationalmarketscan preventthis from happening, and indeed consumptionvolatility can increasewith financialliberalizationin some cases. The argumentdoes not rely on a rise in outputvolatilityresultingfrom liberalization. The contribution most closely related to ours is Attanasio and Rios-Rull (2000). These authorsbuild a model of a village economy in which agents face both aggregateandidiosyncraticrisks. Local arrangementssubjectto limited commitmenthelp agents partlyinsureagainstidiosyncraticrisks. The authorsconsider the consequencesof an outsideprogramthatinsuresthe villagersagainstthe aggregaterisk.They findthatbecauseaggregateinsuranceraisesthe agents'outsideoption, arrangementsto share idiosyncraticrisks deteriorate.The authorsuse this framework to cautionagainstundesirableconsequencesof internationalaid programsto poor village economies in less developed countries.While the model in our paper is methodologically related to this contribution,we address a differentissue and suggest a differentmechanism. In our model, agents use internationalmarketsto smooth both idiosyncraticand aggregate income shocks. Consumptionvolatility increases because of uneven access to foreign marketsand is relatedexplicitly to participationin internationalfinancialmarkets.The frameworkwe use thus allows us to addressthe distributionalaspects of financialintegrationin developingcountries. In addition,the Attanasioand Rios-Rull mechanismdoes not generatean increase in aggregateconsumptionvolatility that can be obtainedin our model. The limited commitmentframeworkhas been applied in internationalmacroeconomics to study risk sharing(see Kehoe and Perri,2002, for a state-of-the-art example).An importantdifferencebetween this paperand existing literatureis that limitedcommitmentis typicallyemployedto model risk sharingbetweencountries. This paperanalyzes a case in which the limited commitmentfrictionaffects agents within a country,and the presenceof internationalmarketsaffects purelydomestic risk-sharingrelationships.

I. An Example Suppose there is an endowmenteconomy populatedby two types of agents,A and B. The agents' endowmentsin each period are composed of an aggregatecomponent common to both groups and an idiosyncraticcomponent, which is perfectly negatively correlated across groups. Suppose for simplicity that the aggregate shock takes on values of rj and -rl with equal probability.TypeA gets an idiosyncratic shock of size o with probability 2, and -o with probability%2, independent of the aggregateshock in that period.Type B's idiosyncraticshock is the opposite of A's in each case. There are then four equiprobablestates of nature,for which agent endowments (eA, eB) are given in Table 1. In this endowmenteconomy with no aggregatesaving,the varianceof aggregate outputis Var(Y)= 4i2. In the closed economy, varianceof aggregateconsumption 'An importantexceptionis an economythatis subjectprimarilyto trend-growthshocks.See Aguiar andGopinath(2004).

241

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Andrei A. Levchenko

Table 1. Agents' Endowment Values by State AggregateState (eA, e

Idiosyncraticstate

s= 1 s=2

High

Low

(1+rl+o,1+r-o)

(1-+0o, 1---w)

(1+rj-O,1+r+O)

(l--o,

1-rj+O)

is Var(C)= 4T"2as well. Assuming agents are perfectly able to shareidiosyncratic risk, the averageconsumptionvarianceacross agentsis Var(c)= 12 Suppose now that this economy opens up to internationalmarkets,but not all agents have access. In particular,assume that only type A can insure abroad,and for simplicity suppose she insures perfectly her income process, both the aggregate and idiosyncraticcomponents.Now A's consumptionis constant,but none of B's risks are insured.Consumptionvalues of agents (cA,cB) are given in Table 2. While in this economy the varianceof aggregateendowmentis still Var(Y)= 4T2, aggregate consumption now has variance Var(C) = Ti2+ 02. The average con-

sumptionvarianceacross agents is Var(c)= 2 (112+ o2). Three conclusions from this simple example are worth highlighting.First, the volatilityof aggregateconsumptiondoes not decreaseunambiguously.It is truethat type A's newfound ability to insure herself against aggregaterisks decreases consumptionvariance.But A's decision to participatein the foreign marketsdeprives type B of the ability to insure her idiosyncraticrisks. Thus, if risks that are insurable withinthe economy are importantrelativeto aggregatecountryrisk, aggregate consumptionvolatility may go up as a result of type A's departure. Second, the aggregatecountryrisk is not eliminatedentirely in this economy. While type A can insure herself against those risks, type B, which does not have access to internationalmarkets,is still subject to aggregate shocks. Thus, when access to internationalmarketsis uneven in this way, the economy may not be able to take full advantageof aggregateinsurancethey provide. Finally, the gains from this type of liberalizationare unevenly distributed.In particular,while A gains from accessing the internationalmarkets,B's consumption volatility increases owing to reducedrisk-sharingopportunitiesat home. This exampleis clearlytrivialized.In particular,the agents'behaviorfollowing openingup to internationalmarketslooks farfromoptimal.Two key questionsarise.

Table 2. Agents' Consumption Values by State, After Opening AggregateState High

(CA, CB)

Idiosyncraticstate

s=1 s=2

(1, 1+r-o) (1, 1+Tl+O+)

242

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Low

(1, 1-r-o) (1, 1-1+0o)

FINANCIAL LIBERALIZATION ANDCONSUMPTION VOLATILITY First,why would type A insureher idiosyncraticrisk abroadwhen she can do so at home? This questionwould be even more relevantif, for example, accessing internationalmarketsis more costly thanaccessing domestic ones, an assumptionmade in the formal model in the next section. Second, why can't type A efficiently pool the entire aggregaterisk of this economy and use her access to foreign insurance technology to insure type B's aggregaterisk as well? In a frictionlessbenchmark that we should keep in mind, agents insure each other perfectly against idiosyncraticrisk before and afterfinancialliberalization,and access to foreign insurance even by a subsetof agents will eliminateall aggregaterisk as well. In the next section we build a model to show that frictions in the domestic risk-sharingsystem lead to outcomes illustratedin this simple reduced-formexample. II. The Model

The Environment Thebasicmodelis a simpleversionof Kocherlakota (1996)or KehoeandLevine (2001).Thereis anendowment economypopulated by twokindsof infinitelylived agents,A andB, withidenticalperiodutilityu(c) anddiscountrateP< 1.Eachgroup hasmass1.Agentsin eachgroupmaximizelifetimeexpectedutility:

t=O

Thereare S statesof nature,denotedby s = 1, .... S, with states occurringwith probabilityPs.Agents' stochasticendowmentsare eAandeBin each s =- 1,..., S. We

makethesimplifyingassumption thatthetwogroupsfaceanidenticalunconditional endowmentprocess.Aggregateendowmentin thiseconomyis Ys= ea+ e4in each states. We assume that endowments are perishable,and so there is no aggregatesav-

ing in the economy.Wheneachagentsimplyconsumesherendowmentin every period,the lifetimeexpectedutilityis givenby: s

~1 Vaut,

EE

t=0

t'u(e )= -

s=1psu(es).

(Here, the subscript"aut"stands for personal autarky.) We model uneven access to internationalmarketsby assumingthatonly agents of typeA can use these marketsto insure.Supposethatthe foreigninsuranceprovides to A in states, s = 1,..., S, for each unitof endowmentthatthe typeA comos type mits to the internationalmarkets.To make the problem interesting, suppose that accessing the internationalmarketshas a cost 7t. Thus, if the type A chooses to insure iy units of her endowmentabroad,she will be able to consume A= eA

Ys

=e, + i4'C)-a4tZ

243

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Andrei A. Levchenko in state s, s = 1,..., S. We make the assumptionthatthe transfersare a pureinsurance, that is, E(ps) = 0. Note that this requiresagentA to transferincome to internationalmarketsin some states (Ps < 0 for those s). The trade-offis clear.If thereis no type B, the optimalforeign marketparticipation will weigh the benefits of insuranceagainstthe costs of buying it, t. When insuranceis costless, the agent insures completely.Access to costly international marketswill now determinethe outside option of the type-Aagent. Let aut,denote the optimalportfolioof internationalinsurancetypeA would choose in the absence of B: Vau, =

argmax

s.t.

psu(y)J

L's=l

'V

~

yaO= ea +

-

Vs =

,

S 1....

W2>0.

We introducethe last constraintbecause of the positive cost of purchasinginsurance,it. Allowing agents to buy negative amountsmay in this formulationlead them to do so for values of nthigh enough, as it can raise their average consumption. We let vAt be the lifetime expected utility type A gets from optimally participating only in the internationalmarkets. Besides the foreign markets, type A can also enter into a risk-sharingrelationship with type B. Domestic risk sharing is subject to limited commitment. Agents can walk away from the relationship at any point. If this happens, the domestic risk-sharingrelationshipbreaksdown forever.We view limited commitment as a consequenceof institutionalimperfectionsin the domestic markets,such as poor contractenforcement.The main problemis that these agents cannot sign a bindingcontractcommittingtheirfutureincome flows to the relationship.The voluntaryparticipationconstraintmust hold in all dates and states, and will limit the amountof risk sharingattainablein this economy. In general, the evolution of risk sharingand consumptionin this economy is history-dependent. Denote by st = { so, ...,

st} the history of the states of nature

throughperiodt. Agents enterthe risk-sharingarrangementby specifyingconsumption allocations (cA(st),CB(st))and foreign marketparticipationby type A, y(st) for each period, and each possible history st, subjectto the participationconstraintof each agent in each date and for all histories,

t

t=t

dj

t=t

and the aggregateresourceconstraintin the economy, 244

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIALLIBERALIZATION AND CONSUMPTIONVOLATILITY

cA+ c,= y + e,

Vt.

(4)

The participationconstraintsstate that any risk-sharingarrangementmust give each agenta lifetime expectedutilitythatis at least as greatas the lifetimeutilitythe agent would get by reneging on the arrangementand consuming her endowment from thatperiodon. The formulationis flexible and incorporatesthe possibilityof a punishment,by introducinga parametery. Whenthereis no enforcementat all, y= 1, andwe arein a worldof no commitment.Wheny < 1, thereis some punishmentthat can be inflictedin case an agentreneges,andthus a widerrangeof risk-sharingrelationshipsare sustainable.We thinkof the parametery as reflectingthe qualityof a country'sinstitutions,with lower values reflectingbetterinstitutionalquality. It is easy to establish the first best benchmark. An allocation (cA(st), cB(st),W(st))is first best if the ratio of marginalutilities u'(cA)/ is constant u'(ct) across time and states, and the economy consumes its full endowment every y + e B,Vt. period, cA + c ytA

RecursiveSolution:TheGeneralCase How can we determinehow much risk sharingand foreign marketparticipation takes place in this economy? While for the most partwe will be comparingsteady states,it is usefulto writedown the generalformulation,to highlightthe most importantfeaturesof the optimalcontractin the presenceof a varyingoutside option for A. TypeA simultaneouslychooses the extentof her participationin the foreignmarkets and the amountof risk sharingthatis takingplace in the domesticrelationship. The recursive representationof the problem above can be obtainedby introducing a state variable,v, which representsthe expected lifetime utility promised to one of the agents, and giving a recursive structureto the participationconstraints.In particular,let v be the utility promisedto agent B and pA(v) be the lifetime utility thatA can attain.pA(v) is given by the following Bellman equation:2 pA(v)

S

max p V) {cs = w,

s.t. pS

I

s=1

)+ u(cs

(P)

o,p[u(Ys s=1

+q~yV-my--Co)+PpAlws)]

s}Sl

+ pw,

(PK)

v

]

y /)+ oPW [u(e ivau, ] -

u(Ys +qs

4

+ --cB)

A (ws)

Vs = 1.....S

[u(eA

(PC") ) s

Aut

VS =

S0 WB W, vm E[vat, E[LS,

1.....S (PCA) (INS)

]1-

S,,J~]

2SeeKocherlakota(1996); andLjungqvistandSargent(2000, ch. 15)

245

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA, Levchenko Equation(P) is theBellmanequationforthevaluefunctionof typeA. Theway the programhas beenset up, typeA choosesforeignmarketparticipation y, consumptionlevels c,B and the expected lifetime utility levels she promises to type B

in eachstate,ws, to maximizelifetimeutilitysubjectto theconstraints. In particular,(PK)is the promise-keeping constraint,whichensuresthattypeB does get theexpectedutilityshe hasbeenpromisedin thepreviousperiod.ThefollowingS constraintsof typeB. Thesearerecursive constraints,(PCB),aretheparticipation of the generalparticipation constraint,equation(3), for eachstate representations of nature.Intuitively,the risk-sharingcontract(cB, Ws) offered to agent B in each

states shouldbe suchthatthe agentis willingto stayin the risk-sharing relationthe outside of her endowment from that shipgiven option consuming periodon. o Theparameter y 1 is meantto measurethequalityof domesticinstitutions.(PCA) are the participation constraintsof typeA. The condition(INS)preventstypeA fromtakingon negativeamountsof foreigninsurance.The last two constraints restrictthe policyfunctionsto the feasibleset. Comparedwiththe canonicalversion of the model, type A's optionto insurein the foreignmarketsintroduces anothercontrolvariable,the optimalforeignmarketparticipationy. Let ,i be the multiplieron thepromise-keeping constraint(PK);Psksthe mulconstraintsfor B, (PCB);ps Osthe multipliers tiplierson eachof the participation on each of the participation constraintsforA, (PCA);and8 the multiplieron the constraint(INS).Thenthe first-orderconditionswith respectto J-nonnegativity Cs,Ws,s

=

1 ... , S, and y are

-(1+e,)u'(Ys +cpsy(1 +O,)PA

(Ws

)+(

+s)=

y --cf )+(Ct+,

)u'(c,

)=O

Vs=l ...,S

Vs= 1,... S

0

(5) (6)

s= S

+p, Po,

r)[u'(Y1 +

-

-

, (, -)[(,

+s=1 + 6=0o.

-

P,w-v-C)-'(e

c)

-

(7) +p.-r)]

By the envelopetheorem, P'(v)=

.

The first two first-orderconditions can be combined to yield the optimalrelationshipbetweenconsumption givento B andpromisedutility,in eachstate: ul(Ys + (PW- 7cy,- cB) Uo(CB)

P'(w,).

(8)

andSargent(2000),we observethatthere Followingthediscussionin Ljungqvist

are three kindsof states. If in states neither(PCA)nor (PCB)binds,?, = Os= 0, ws = v, PA(ws)= pA(v), and the values of consumptionare solved from equation(8). 246

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIAL LIBERALIZATION ANDCONSUMPTION VOLATILITY In stateswhere (PCB)binds, ks> 0, ws > v, PA(Ws)< pA(v),agentB's promisedutility increases,andA's lifetimeutilitydecreasesas a result.In this state,c s and can be obtainedby solving equation(8), and (PCB)holding with equality,for aws given equilibriumvalue of y. The oppositeis truefor statesin which (PCA)binds.In those states, ws < v, and pA(ws) > pA(v).

The optimalparticipationin foreign markets,W,is determinedby equation(7). Though it is a complicatedexpression, it containsthree distinct partsfrom which we can glean some intuitionfor what drives the choice of x. The first term is the "optimalportfolio"term. It tradesoff the benefits of insuranceagainst its cost it, and it would be presentwhetheror not participationconstraintsforA bind.The second termcomes from the effect of portfoliochoice on A's participationconstraints. In particular,if in any states A's participationconstraintbinds (thatis, 6, > 0), agent A will take into account the effect of foreign insuranceon her participationconstraintin that state. Note that we cannot tell in the general case whetherraising y relaxes or tightens the constraint;thus, the effect of the presence of these constraintson the equilibriumamountof foreignparticipationis ambiguous.In the specific two-stateexamples we work out below, however, the intuitionfor this effect will be quite clear.The thirdterm simply comes from the nonnegativityconstraint we imposed on the foreign marketparticipation. Kocherlakota(1996, Propositions4.1 and 4.2) shows that startingfrom an initial value of vo for which nontrivialrisk sharingis possible, the relationshipconverges to a steady state, in which the first best level of risk sharingmay or may not be attained.Unfortunately,an analyticsolutionto the Bellmanequationin (P) is not known even in the canonicalversion of the model, which does not includeendogenous foreign marketparticipation. To understandhow the equilibriumamountof risk sharingrespondsto changing opportunitiesto participatein foreign markets,we will assumefunctionalforms and solve for the value and policy functionsnumerically.In all cases we consider,a value functioniterationmechanismdescribedin Judd(1998, ch. 12) straightforward is sufficientto generatea solution. We approachthe problemby consideringthe two extremecases, those of purely idiosyncraticand purelyaggregaterisk. Looking at simple versionsof this problem lets us gain a fairbit more intuitionaboutthe effect of financialliberalizationin this environment.It also allows us to reducethe numberof states to the minimumpossible value of 2, thereby significantly reducing the dimensionalityof the policy function.

Case I: PurelyIdiosyncraticRisk We now consider the first of the two polar cases. For simplicity, suppose there are two statesof nature,s = 1, 2, and the stateshave equal probabilityof 2. When there is no aggregaterisk, agents'incomes are perfectlynegativelycorrelated.In particuA lar, we assume that in s = 1, groupA's per capitaincome endowmentis e = 1 + = 1 - E. In s = 2, the per capita endowements are reversed.The total endowmentin the economy equals 2 in every period. and group B's per capita endowment is

247

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA. Levchenko The foreign insurance provides -pf to type A in s = 1, and pf in s = 2, for each

unit of endowmentthatthe type A commits to the internationalmarkets.If the type A chooses to insure a share i of her endowmentrisk abroad,she will be able to consume y=

-

1+ E-

o

in s = 1, and A y=1-

+ (P

-

in s = 2. It is useful to restatethe recursiveformulationfor this special case, writingout participationconstraintsstate by state:

PA(v)=max

{=

PA(v)

c~

2-

pA

,{cw}s,1,2

s.t.

fw

u(cf

2 y[u(12 y[u(l

)-+ow

E)+ +E)+

]o v ] pvaut, vaut]

cpf- sy - ci)+PPA

w)B-uy[u(1]

u(2+-

0 S>_ cB" E [0,Y, i'] + s w,

(P)

c)+P

1 2-[u(cfl)+ w +u(c )+fw u(cj)+

sB (wfB)]

(PK') (PC')

(PC')B/

(PC)A/ (INS)

e[Vaut,,mVx].

This formulationis quite general and includes a numberof importantspecial cases. The closed-economy case is replicatedwhen rt is prohibitivelyhigh, so that even without type B, type A would not want to access the internationalmarkets (Naut = 0). Then, vAut = Vaut, and the domestic risk-sharingrelationshipis intact. Another importantspecial case is that of frictionless domestic marketsgiven by y= -oo: the participationconstraints never bind, and the first best outcome is achieved. At anotherextreme, suppose that there is no commitment,y = 1, and internationalmarketsare costless (t = 0). Then, we know that Naut = (without B, type A opts for full insurance).Under these circumstances,the domestic risksharingrelationshipwill most likely break down completely, because the type B agents would not be able to providetype A with favorableenough termsof domestic insurancewithout violating their own participationconstraint.If domestic risk sharingbreaksdown, type B is completelyuninsured.The discussionof the extreme

248

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIAL LIBERALIZATION ANDCONSUMPTION VOLATILITY cases provides an illustrationthat domestic risk sharingis likely to sufferthe most when the cost of accessingforeignmarketsis low anddomesticinstitutionsarepoor. Even before going to a numericalsolution,we can make two importantremarkson the featuresof this problem. Remark 1: If the first best is reached in this risk-sharingcontract,it necesof the sarily means thatthereis no foreign marketparticipation, =0, - irrespective value of it. Owing to the absence of aggregateuncertainty,the first best level of risk sharingimplies thatall agents'consumptionis constantacross time and states: agents are perfectly insured. Since risks are perfectly insurablewithin the economy, in the frictionless setting there is no role for internationalmarketsin smoothing consumptionrisk. Participationin internationalmarketsreduces welfare in two ways, vis-a-vis the first best benchmark.First, it costs n, and it thus reducesthe aggregateendowment in both states. Second, and most importantly,because the agents' endowments are negatively correlated, type A's participationin the foreign markets actually lowers her ability to insure type B. In particular,whereas in the closed economy A had at her disposal Ein s = 1, with which to insureB's negative income shock of -E, now in s = 1 agent A has only E - pfl - niy. The feature that the first

best benchmarkis the same for each t is also convenient because as we consider the effects of financial liberalizationon domestic risk sharing,we can judge the changing amountof domestic risk sharingagainst a constantbenchmark. Remark 2: When in equilibriumthe amountof foreign participationis y = 0 and the first best level of risk sharingis not achieved,lowering barriersto internationalmarkets,i, actuallydecreasestypeA's welfarepA(v),for each v.This is a consequenceof the envelope theorem.Evaluatedat an optimumvalue of S S==O,

= dp

P s

sdP(v) s=l

dat

How can lowering the internationalbarrierstype A faces make A worse off? Internationalmarketsplay two roles in our framework.First, insuringabroadmay improveA's lifetime utility by smoothing some of A's consumptionrisk. Second, ability to access internationalmarkets raises A's outside option, irrespective of whetherA actually participatesin the internationalmarkets.The second effect is detrimentalto A's ability to insure domestically.Thus, if there are parametervalues under which A chooses not to insure abroadat all (i = 0), only the second effect remains.By raisingA's outside option, the presenceof foreign marketsactually decreases the amountof risk sharing attainablein the domestic relationship, lowering A's utility for a given v. The pureidiosyncraticrisk economy in this subsectionprovides the most drastic illustrationof the perverse effects of internationalmarkets on domestic risk sharing.Though in the first best world internationalmarketshave no role, in the limited commitmentframeworktheirmere presencehas a negative effect. Remark 2 focuses on A's participationconstraints,but B's constraintsmatteras well. Since

249

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Andrei A. Levchenko A's insurancein foreign marketsdecreases aggregatewelfare, type B has an incentive to induceA to lower her foreign marketparticipation.The ability of type B to offer A better domestic risk-sharingterms is limited, however, by type B's own participationconstraints.There is only a limited amountof utility that B can give up before they startto bind. We now provide a numericalillustrationof the effect of financial opening on agents' welfare and consumption volatility. To do this, we assume a functional form for the utility function that is quadratic: 1

u(c) =4c - -c2 2

The parametervalues we pick are the following: o = 1,
V1.

As we lower it, we see thatthe frontiershifts unevenlyinward.In particular,two key observationscan be madefrom this figure.First,the Paretofrontieris no longer symmetric.The pairs (v, PA(v)) of sustainablelifetime utilities become skewed in favor ofA: if PA(v1)= v2, then PA(v2)> V1.Second, the range of values of v for which

nontrivialdomesticrisk sharingis sustainableshrinksas we lower internationalbarriers.This is intuitive:the higherA's outside optionbecomes, the lower is the maximum value of B's lifetime utility v for which A is willing to participatein the domesticrisk-sharingrelationship.We also see thatfor each v, the lifetime utility of A, pA(v), decreases in ntin this example, as long as in is high enough to sustain domesticrisk sharing-an illustrationof Remark2. While finding the value functionpA(v) is informativeabout the combinations of the two agents' lifetime utilities that are sustainablein the economy, it does not tell us much directly about the amountof risk sharingand foreign marketparticipation that occurs as t changes. We can perform comparativestatics by finding steady-state levels of risk sharing and foreign marketparticipationfor different values of it.

In a steady state, income transfers,and thus consumption,are constant over time in each state, though not necessarily constant across states (see Kehoe and Levine, 2001, Proposition5). It is straightforwardto show that in a steady state, expected lifetime utility,denotedby vi, i = A, B, is constantfor each agent as well. We can fully characterizethe symmetric steady state by consumptionvalues of each agent in each state, {CA,

,

Ca,

?}.

We label steady-state values with an

overbar. The key limitationto the extent of risk sharingthattakes place in this economy is the voluntaryparticipationconstraintthatmustbe satisfiedfor each agentin each 250

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AND CONSUMPTIONVOLATILITY FINANCIALLIBERALIZATION

Figure 1. P(v) for DifferentValues of n, with PurelyIdiosyncratic Risk 20

19

18

P(v) 17

Firstbest 16

it

=0.125\ TC=0.15\

\450

15 15

IC

=0O2

Closed'economy 17

16

18

19

20

V state and each period. In practice, risk sharing takes place by transferring income from the group that has a high current income realization to the other group. Naturally, then, the only relevant participation constraints will be those in which the current realization of income is high for that particular group. There are two possibilities. If in steady state, the participation constraint of the agent that is experiencing a high income shock does not bind,

u(c,

)+ ivA >

u(c2+ 2

u(c )+

+ + 'y(u(1 e) ov^,),

(u(+E)+pv),

9Au

>o

then the first best level of risk sharingis achieved, and each agent's consumption is constant across time. Notice that in this type of steady state no participationin the internationalmarketstakes place. 251

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA. Levchenko If, on the other hand, in steady state the participationconstraintsbind, the steady-stateconsumptionvalues {RA,c-, cB, co} are those thatmaximize VA,subject to participationconstraintsholding with equality: -A

-A oov

-B

"o

)+ pvA ),

-o(PfV

-B

u(c2 )

-

y(u(1 + E

u(c )+p

) = o/U(

d-

"o

1aut

*.

We illustratehow the steady-stateamount of risk sharingchanges as barriers to accessing the foreign markets, rt,are lowered. Here, we consider the same set of parametervalues we used to constructPA(v)above. At these parametervalues, the closed economy does not achieve perfect risk sharing,and the steady state is unique. The effects we discuss are much more general, however. Figure 2 illustratesthe patternsof consumptionfor the two types in the two states as a function of the cost of accessing internationalmarkets, t. Solid lines represent consumption values of type A and dashed lines of type B. Without domestic or internationalrisk sharing,each type would consume her endowment, which is equal to 2 in the high state and 0 in the low state. Perfectrisk sharing,on the otherhand, implies thatin a symmetricsteady state, consumptionis equal to 1 for all agents in all states.

Figure2. Steady-StateValuesof Consumptionas a Functionof Access to ForeignMarkets,withPurelyIdiosyncraticRisk 2.5

2

1.5

C

-B C2 -A Cl

1

0.5

-A dk C2

-B

Cl

0

~If -0.5

7C3

7r2

711

7C

252

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIAL ANDCONSUMPTION LIBERALIZATION VOLATILITY How does the option of accessing the internationalmarketsaffect risk sharing at home? We can divide values of it into four intervals. First, when the cost of accessing the internationalmarketsis prohibitive,i > irtl, they to not raise type A's outside option, aut = 0, and vut = vaut. The foreign marketsare too expensive, and, even if left alone, type A would choose not to participatein them. In this case, risk sharingis the same as in the closed economy. When nt < itl, the presence of foreign markets does raise type A's outside option, because Waut, > 0. When it2 < t < it1, the outside option of type A is rising, but foreign marketsare costly enough that type B can induce A to stay entirelyin the domestic risk-sharingarrangement.Notice that as t decreases and the outside option of type A rises, the amountof risk sharingtaking place decreasesfor both agents, but type A's consumptionis higher in both states than the corresponding consumptionof type B. This is because the rising outside option for A both reduces the amountof risk sharingavailableto agents and increases the transferof utility that type B must make to keep type A at home. In this interval, type A does not participatein the foreign markets,y = 0. Thus, while there is less risk sharingat home, aggregateconsumptionis still flat. When it falls below it2, some foreign marketparticipationstartsto occur. As some of the type A's consumptionrisk is now insured abroad,her consumption volatility startsdecreasing.But this also means that there is less possibility of risk sharingat home, and consumptionvolatility of type B continuesrising. This is precisely the effect illustratedin the simple example of the previous section. While participationin internationalmarketscan decreaseconsumptionvolatility of some agents, it can have adverse effects on consumptionvolatility of others. Type A's rising participationin foreign marketsimplies that she is less able and willing to insure type B. Finally,when Rt2 it3,internationalmarketsare so accessible, and thus type A's outside option is so high, that type B cannot offer good enough termsof insurance contract at home without violating her own participationconstraint. Thus, all domestic risk sharing breaks down, and type A participatesonly in the international markets.The problemwith this, of course, is that type B is now completely uninsured.Aggregate consumptionvolatility is highest, and the type B agents are least insured,when opening up to internationalmarketsimplies a complete breakdown of domestic risk sharing.

Case II:Aggregate Risk Supposeinsteadthatall agentshave identicalendowmentsin each period,in particular, in s = 1, eA = e - 1 + e, and in s = 2, eo = eB - 1 - E. Notice that when the economy is closed, there is no scope for risk sharing.When the economy is open, the first best allocationrequiresthattype A pool the entirecountryrisk and insureit optimallyin the foreign markets,given the cost of access it. The internationalmarkets are modeledexactly the same as in the previoussubsection,transferring-ipf in s = 1 and qpfin s = 2 for each unit of endowmentinsuredabroad.

253

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Andrei A. Levchenko When the relationshipbetweentypes A andB is subjectto limitedcommitment, we can characterizeit by setting up a programsimilarto that of the previous subsection, as a value maximizationof type A, PA(v), subjectto constraints: 2

PA

p(v. (v)s=BmaxB mawB }s=1,2

=

2t[u

+ p

-

- C) +

[pa

(p")

(Wf)]

Ps)

s.t. Pwi +u(cfl)+ 1[u(ci8)+ 2

wfl]

>

(PK")

v

u(c.f) +p w 2 y[u(l1+ E)+ B1~(PCK") ovau,], u(c.f)+ w 2 y [u(1- E)++v,,, u (2 -

- c

cpfI

.t

u(2 + pfy -

o0 >

ny

(PC2B/ II\ )

)+

PP A (w B) y U( + E B > -- cf ) + PPA (2 ) > yfU(1-- E +

SE [,+ Vtm +sV -

(PC2")

--

+

, VAt

(pY--T)VaAut

(

'" 2A

) (INS")

]

W, E[Vaut, Vmax].

Examiningthe constraintsallows us to get a sense of what limits efficient risk pooling in this economy. The participationconstraintof type B in the high state (PC ") shows that ratherthan transferringincome to type A for the purposes of insurancein the internationalmarkets,type B will be temptedto consume her current high endowment, an intuition identical to that of the previous subsection. When the economy is experiencinga negative aggregateshock, type A is the only one with access to a net transferfrom abroad.Efficient risk pooling would call on type A to redistributesome of the positive income to type B, but that is limited by A's participationconstraintin this state of nature,(PCA"). It is importantto note that the relationshipbetween types A and B is very different here comparedwith the idiosyncraticrisk case. In that case, domestic and foreign insurancewere substitutesfor type A. Here, engaging with type B serves no insurancepurpose for type A, and to induce type A to take on a risk-pooling role, type B must transferincome to type A. Type B's ability to decrease her own utility in the risk-poolingarrangementis itself limited by her voluntaryparticipation constraint. Once again we use a numericalexample to provide an illustration.We could repeatthe exercisein the previoussection andlook at the responseof riskpooling to changingvalues of it. However,when the economy is subjectto aggregaterisk, the first best frontierchanges as we vary t; thus, we don't have a naturalbenchmark. Instead, we use this example to highlight the importanceof quality of contract enforcement,y, in determiningthe amountof riskpooling achievedin this economy. For simplicity,we assume that thereare no barriersto internationalmarkets, t = 0. The first best in this economy is achievedby sharingall of the aggregaterisk in the internationalmarketsand by giving each agent constantconsumptionacross time. 254

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AND CONSUMPTION VOLATILITY FINANCIAL LIBERALIZATION

Noticethatthefirstbestfrontierin thiseconomyis thesameas in theidiosyncratic risk case of the previous subsection,but it is achievedvery differently,throughfull

in international markets. participation How do institutionsaffectthe amountof risk sharingachievedin this economy? Figure 3 plots the first best frontierand the value functions, pA(V),for seVeral values of y. Several aspects of this figure are worthhighlighting.First, better institutionsimply that the economy is closer to the first best frontier.As institutionsget worse,thefrontiershiftsinward.ThisimpliesboththatpA(V)is lowerfor a given v, and that the range of B's lifetime utilities, v, for which nontrivialrisk sharing is attainable, is narrower.For high enough y (in this example, about 0.8875), no risk pooling is possible, and type A insures in the internationalmarkets alone, leaving B completely uninsured.Second, the figure illustratesthe distributionalconsequences of uneven access to the internationalmarkets.When A can insure abroad,she must be given lifetime utility at least as great as what she would get from perfect insurance abroad (17.5 in this case, given by a dashed

Figure 3. P(v) for Different Values of y, with Purely Aggregate

Risk

20

19-

18-

y=0.875t

P(v) 17-

y=0.8625

y=0.8

y =0.8375

Firstbest

16-

15 15

16

17

18

19

20

V 255

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

Andrei A, Levchenko Figure 4. Steady-State Values of Consumption as a Function of Domestic Institutions, with Purely Aggregate Risk 2.5

-B Cl-. 2 W/

CA iC2

1.5

1-

C

-A Cl

0.5

BC2

0

-0.5

I

I

I

line). This-necessarilymeans that,as B engages A in an insurancerelationship,B's lifetime utility is smallerthanA's. It's importantto note that this statementis true whetheror not the economy achieves an allocation that is first best. To give a sharperpictureof the amountof insuranceagentsget in this arrangement as a functionof y, we can comparesteady states in this economy in a manner similar to the previous subsection.Figure 4 plots the steady-statevalues of consumption for different levels of institutionalquality. There are several distinct insurancerelationshipsthat can arise, dependingon the value of 7. Startingat the left-handside of the graph,when y < institutionsare strongenough that a risky1, pooling contractunderwhich both agents are perfectly insuredand receive equal lifetime utility is sustainable.As we move into the intervaly e (71,Y2), the riskpooling relationshipcan no longer sustain an equitable allocation. In this area, aggregaterisk is still perfectlyinsuredby the economy,andbothagentsareperfectly insured.But to induce A to performthe risk-poolingrole, B must give up utility. Thus, while both agents' consumptionis constantacross time, A's consumptionis higherthanB's.3As we move into the intervalye (72,73),imperfectinstitutionsprevent efficient foreign insurance,even in aggregate.Neither agent is now perfectly insured, but the risk-pooling relationshipstill operates, and A provides positive insuranceto B. As we lower institutionalqualityin this interval,agents are less and 3Generally,whenthe economyattainsa firstbest allocation,as it does in this interval,the steadystate is not unique.In constructingthis figure,we select for eachy the steadystatein whichthe two types'consumptionvaluesareclosest to each other-the mostequitablesteadystate.

256

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIAL LIBERALIZATION ANDCONSUMPTION VOLATILITY less well insured.Finally,when y> 73,no risk pooling is sustainablein equilibrium. This means thatA leaves the domestic relationshipentirely and insures optimally (perfectly)abroad.It also means that B is completely uninsuredand consumes her own endowmentin each period. To summarize, the first best outcome of efficient risk pooling may not be achieved in this economy. When risks are purelyaggregate,access to international marketsimprovesthe economy's aggregateconsumptionvolatility.TypeA certainly reachesthe optimal level of insurance,given the cost of access it. The benefits of financialopening may not spreadto agents who do not have direct access to internationalmarkets.

Ill. Conclusion The latest empiricalevidence demonstratesthat increasinginternationalfinancial integrationis actuallyassociatedwith higherconsumptionvolatility in developing countries. This finding is difficult to rationalize within the frameworkof representative agent models of risk sharing. The main shortcomingof representativeagent models is that they can tell us only about the role of internationalfinancialmarketsin sharingaggregatecountry risk. The canonical models also do not addressthe issue of how aggregaterisk is pooled amongagentsfor the purposesof internationalrisksharing.This papershows thatfocusingon agentheterogeneitywhen domesticrisk sharingandriskpoolingare subjectto frictionscan help us rationalizethe empiricalevidence. In the model we presented,agents are heterogeneousin both their income process and in whether they have access to internationalmarkets.When we considerthe consequencesof agent heterogeneityin an economy with underdevelopedinstitutionsand financial markets,we reachthreemain conclusions. First, if income risks are idiosyncratic,financial opening will have first-order effects on the domestic financial markets.Internationalmarketscan be used by agentsto insureagainstnot only aggregatebut also idiosyncraticrisks.When some agents participatein the internationalrisk-sharingmarkets,domestic risk sharing deteriorates,leading to an increase in consumptionvolatility.The mechanismwe suggest here reproducesa positive relationshipbetweencapitalflows andconsumption volatilityfound in the data. Second, when agents face only aggregate income risk, the underdeveloped financialsystem will preventefficient pooling of risk acrossagentsfor the purposes of internationalinsurance.Thus, for aggregaterisk, the benefits of access to internationalmarketsare much lower in this frameworkthanin the representativeagent model. Finally, considering agent heterogeneityallows us to highlight distributional consequencesof financial liberalization.While agents with access to international marketsbenefit from expandedopportunities,those who do not have access benefit less, and in fact may experiencean increasein consumptionvolatilityand a reduction in welfare. Can we find empiricalevidence corroboratingthe mechanismwe describehere? Our focus on the quality of the financial system and institutionsreceives some 257

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

AndreiA. Levchenko empiricalsupportin Bekaert,Harvey,and Lundblad(2004), who show that financial liberalizationleads to an increasein consumptionvolatilityin countrieswith the least developed financial systems and low institutionalquality. In contrast,consumptionvolatilityfell afterfinancialliberalizationin countrieswith well-developed financialsystems and good institutions.These resultssuggest thatfocusing on institutionsas we do in this paperis a fruitfuldirection,but they are admittedlyuninformative aboutthe precise channelfor the effect. More importantly,is the mechanismwe are proposingplausible?In our model, aggregate consumptionvolatility can increase after financial liberalizationwhen idiosyncraticshocks are large enough relativeto both the size of the economy and the aggregateshock that,when left uninsured,will impactaggregateconsumption. Clearly,such shocks do not correspondto iid income shocks to individualatomistic consumersor firms,which would averageout in a countrypopulatedby a very large numberof such agents.The idiosyncraticshocks in our model are betterinterpreted as shocks to importantsectors in the economy, or perhapsshocks to large firms. Gabaix(2004) shows thatwhen economies aredominatedby small numbersof very largefirms-as appearsto be the case in practice-firm-specific shocks will lead to aggregate fluctuations.While the line of research that focuses on idiosyncratic shocks to big agents is broadlyconsistentwith the kinds of effects we model here, it still remainsin its infancy. The simpleframeworkwe presentedin thispaperfocuses narrowlyon the opportunitiesfor insuringincome risk domesticallyand internationally.Clearly,financial liberalizationhas a varietyof othereffects on developing economies. Not the least important,for instance, is the role of capital flows in generating output growth throughtheir ability to mobilize foreign savings for domestic investment,or their role in technology transfer.The effects we reveal here shouldneverthelessbe taken into accountin buildinga complete pictureof financialliberalizationin developing countries.

REFERENCES Acemoglu,Daron,andFabrizioZilibotti,1997, "WasPrometheusUnboundby Chance?Risk, Diversification,and Growth,"Journalof PoliticalEconomy,Vol. 105, No. 4, pp. 709-51. Aguiar,Mark,andGitaGopinath,2004, "EmergingMarketBusinessCycles:The Cycle Is the Trend"(unpublished;Chicago:Universityof Chicago). Attanasio,Orazio,andStevenDavis, 1996, "RelativeWageMovementsandthe Distributionof Consumption," Journal of Political Economy, Vol. 104, No. 6, pp. 1227-62.

Attanasio,Orazio,and Jose-VictorRios Rull, 2000, "Consumption Smoothingand Extended Families" in Advances in Economic Theory: Eighth World Congress, ed. by M. Dewatripont,L. Hansen,andS. Turnovski(Cambridge,U.K.:CambridgeUniversityPress). Baxter, Marianne,and Mario Crucini, 1995, "BusinessCycles and the Asset Structureof Foreign Trade," International Economic Review, Vol. 36, No. 4, pp. 821-54.

Bekaert,Geert,CampbellR. Harvey,and ChristianLundblad,2004, "GrowthVolatilityand FinancialLiberalization," NBER WorkingPaperNo. 10560 (Cambridge,Massachusetts: NationalBureauof EconomicResearch).

258

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions

FINANCIALLIBERALIZATION AND CONSUMPTIONVOLATILITY

Calvo, Guillermo,1998, "Varietiesof Capital-Market Crises,"in The Debt Burdenand Its ed. G. Calvo and M. King,IEAConferenceVolume, for Monetary Policy, Consequences by No. 118 (New York:St. Martin'sPress). Gabaix, Xavier, 2004, "PowerLaws and the GranularOrigins of AggregateFluctuations" (unpublished;Cambridge,Massachusetts:MassachusettsInstituteof Technology). Hayashi,Fumio, JosephAntonji,and LaurenceKotlikoff,1996, "Risk-SharingBetween and WithinFamilies,"Econometrica,Vol. 64, No. 2, pp. 261-94. Massachusetts: MITPress). Judd,Kenneth,1998,NumericalMethodsin Economics(Cambridge, Kaminsky,Graciela,andCarmenReinhart,1999,"TheTwinCrises:The Causesof Bankingand Problems,"AmericanEconomicReview,Vol. 89 (June),pp. 473-500. Balance-of-Payments Kehoe, Timothy,and David Levine, 2001, "LiquidityConstrainedVersusDebt Constrained Markets,"Econometrica,Vol. 69, No. 3, pp. 575-98. BusinessCycles with EndogenousInKehoe,Patrick,and FabrizioPerri,2002, "International Vol.70, No. 3, pp.907-28. Econometrica, completeMarkets," of EfficientRisk SharingWithoutCommitment," Kocherlakota, Narayana,1996, "Implications Reviewof EconomicStudies,Vol. 63, No. 4, pp. 595-609. Kose, M. Ayhan,EswarPrasad,and MarcoTerrones,2003, "FinancialIntegrationandMacroIMFStaffPapers,Vol.50, SpecialIssue,pp. 119-42. economicVolatility," Ljungqvist,Lars,andThomasJ. Sargent,2000, RecursiveMacroeconomicTheory(Cambridge, Massachusetts:MITPress). GlobalDiversification, andGrowth," AmericanEconomic Obstfeld,Maurice,1994,"Risk-Taking, Review,Vol.84 (December),pp. 1310-29. and KennethRogoff, 2000, "TheSix MajorPuzzlesin Macroeconomics:Is Therea CommonCause?"in NBERMacroeconomicsAnnual,ed. by Ben Bernankeand Kenneth Rogoff (Cambridge,Massachusetts:NationalBureauof EconomicResearch.

259

This content downloaded from 185.2.32.106 on Wed, 25 Jun 2014 05:06:44 AM All use subject to JSTOR Terms and Conditions